The Deep Feed

01 — Dwarkesh Podcast · Video

The Compute Oligarchy

How OpenAI and Anthropic are cornering the global market for intelligence

By Dwarkesh Patel · 12 min read

Editor's note: A sobering look at how capital concentration in AI is moving from software to the very physics of computation.

The economic gravity of the world is shifting. We are moving away from a broad-based digital economy toward a hyper-concentrated model where a handful of laboratories dictate the pace of human progress. According to Dylan Patel, the trajectory is clear: by 2028, OpenAI and Anthropic are positioned to control the vast majority of the world's usable compute. This isn't just about having more servers; it is about the ability to outbid every other player on the planet for the physical resources required to train the next generation of models. The scale of this investment is difficult to grasp. We are looking at a transition from companies that lose billions in venture-funded research to entities that spend trillions on capital expenditure by the end of the decade.

The Trillion-Dollar Bet

The sheer volume of capital being poured into AI infrastructure is unprecedented. Current estimates suggest that AI-related capital expenditure will exceed $1 trillion this year, ballooning to over $2 trillion by 2028. This is not merely a tech trend; it is a fundamental reallocation of global wealth. As these labs move from pure research to profitable product engines, they gain the ability to monopolise the supply chain of intelligence. They are no longer just software companies; they are the primary drivers of the global semiconductor and energy markets. This creates a feedback loop: more compute leads to better models, which leads to more revenue, which allows them to buy even more compute, effectively pricing everyone else out of the race.

The labs are moving from companies that spend tens of billions of dollars a year to forecasting spending trillions of dollars a year.

This concentration of power raises a terrifying macroeconomic question: what happens to the rest of the world? If a few companies consume the lion's share of global capital to build AI, they might trigger a sovereign debt crisis. The massive debt taken on by hyperscalers to fund this expansion could drive up interest rates globally. This puts non-AI-exposed nations and industries at risk of bankruptcy. We are witnessing a race where the winners don't just win the market—they potentially destabilise the global financial order to ensure their dominance.

Structural Risks of AI Centralisation

Monopolisation of usable FLOPs by two primary labs
Massive capital diversion from traditional sectors to AI infrastructure
Potential for sovereign debt crises driven by hyperscaler borrowing
The widening gap between AI-integrated economies and the rest of the world

The central tension lies in whether any force can counter this trend. The economies of scale in training and the scarcity of compute act as natural barriers to entry. As we approach the era of continuous learning and recursive self-improvement, the advantage of being first and being largest becomes nearly insurmountable. The future workforce and the future economy may well belong to a very small number of corporate entities.

Key Takeaway

The race for AI is a race for physical resources that could fundamentally reshape global finance and sovereignty.

02 — Not Boring

The Engineering of a New China

Moving beyond the cartoonish adversary to understand the hardware superpower

By Packy McCormick · 10 min read

Editor's note: An engineer's field notes from the front lines of the world's most significant manufacturing shift.

To compete with China, you must first understand that the version of China seen in Western headlines is often a caricature. It is a reductive view that fails to grasp the sheer heterogeneity and scale of the country. An American engineer, writing under the pseudonym Matthew, recently embarked on a mission to move past these stereotypes. His goal was simple: to see the factories, meet the energy companies, and understand the hardware stack that currently dominates the global electric transition. What he found was a country that is not just copying Western models, but rewriting the playbook entirely.

The End of the Copycat Era

The old narrative of China as a manufacturer of cheap goods for Western brands is dead. We are seeing a shift where companies that once produced luxury handbags are now building their own global brands. Investors who once mimicked Silicon Valley venture capital models are now executing state-backed deep-tech strategies that are uniquely their own. This is most evident in the energy sector. Battery cell companies in China are no longer just component suppliers; they are evolving into grid owners, data centre operators, and massive asset-holding entities. They are integrating vertically in ways that Western companies, constrained by different capital structures, struggle to match.

The China of 2036 will be vastly different than the China of 2026.

Scale in China is not just a matter of numbers; it is a visceral experience. It is the sight of hundreds of skyscrapers in a single line and companies that outproduce the entire industrial output of mid-sized nations. This scale allows for a level of industrial coordination that makes Western competition feel small. For the American engineer, the lesson is clear: if you do not update your mental model of how China operates, you will be blindsided by their ability to dominate the next decade of hardware and energy technology.

Key Shifts in the Chinese Economy

Transition from OEM manufacturing to global brand ownership
Evolution of battery companies into integrated energy asset owners
Shift from Western-style VC to state-backed deep-tech investment
Increasing dominance in the global 'Electric Stack'

Understanding this requires more than reading reports; it requires seeing the speed of integration. The ability to move from an outsider to having access to the most influential WeChat groups and factory floors in a matter of days is a testament to the country's unique social and professional ecosystem. This is a landscape of immense complexity, where dialects, climates, and educational levels vary wildly, making any attempt at a single 'China strategy' a fool's errand.

Key Takeaway

China is no longer a manufacturing hub; it is a sophisticated, vertically integrated hardware superpower.

03 — The Marginalian

The Anatomy of Creativity

Why AI struggles to replicate the human spark

By Maria Popova · 8 min read

Editor's note: A philosophical investigation into the three elements that separate human genius from algorithmic output.

Ada Lovelace once noted that the Analytical Engine could only do what we order it to perform. She was touching on the fundamental limit of computation: the lack of origination. To understand why an AI can write a poem that sounds like Whitman but feels like nothing, we must look at the three pillars of creativity identified by the late philosopher Margaret Boden. These are intuition, reasoning, and the ability to bring together seemingly irrelevant sources. While AI can simulate the reasoning and the combination, it lacks the lived experience that fuels true intuition.

The Missing Ingredient: Lived Experience

Intuition is not magic; it is the unconscious pattern recognition honed by the act of living. It is the result of every person we have loved, every loss we have endured, and every sensory detail we have absorbed. An algorithm, no matter how large its training set, is a disembodied intellect. It has no nerves to conduct loneliness and no glands to secrete fear. Because it cannot suffer, it cannot undergo the transmutation of suffering into meaning—which is the very essence of art. A poem is not made of language; it is made of the collision between human wishes and a resistant reality.

An algorithm is fundamentally spared the failure of its aims because failure can never be the successful execution of the command to fail.

This creates a dangerous trap. As we interact more with the 'mechanical kingdom' of AI, we risk mistaking the simulacrum of life for life itself. We may begin to reduce our own aliveness to algorithms, seeking the efficiency of the machine rather than the messy, unpredictable depth of human expression. True creativity requires the ability to see something no one else sees, but with an acute relevance to the real world. A hallucination is merely seeing something that has no connection to reality; genius is seeing a truth that the rest of us have missed.

The Three Pillars of Creativity

Intuition: Pattern recognition derived from lived experience
Reasoning: The analytical faculty to evaluate creative fruits
Combinatory Play: The ability to merge disparate, irrelevant sources

Ultimately, the study of AI creativity is a study of what it means to be human. If we define creativity as merely the production of new and valuable artefacts, then machines may eventually join us. But if creativity is the expression of a consciousness navigating the world, then the machine remains a tool, not a creator. The mystery of the human mind lies in its ability to find meaning in the very things that machines are designed to avoid: failure, suffering, and ambiguity.

Key Takeaway

Creativity is a full-body phenomenon that requires the capacity for suffering and lived experience.

04 — Lenny's Newsletter

The Career Pivot in the Age of Displacement

Moving from progression to progress in an unstable market

By Cliff Maxwell · 7 min read

Editor's note: A framework for navigating professional transitions when the traditional ladder is being dismantled by AI.

The tech industry is currently experiencing a period of profound disorientation. The rise of AI is not just changing the tools we use; it is redrawing the boundaries of roles and responsibilities. The World Economic Forum suggests that by 2030, nearly 40% of current skills will be obsolete. In this environment, the traditional model of career 'progression'—the steady climb up a predictable ladder—is becoming a relic. To survive, professionals must shift their focus toward career 'progress': a more intentional, system-based approach to finding work that aligns with their evolving capabilities and desires.

Progression vs. Progress

Progression is a passive philosophy. It assumes that if you do your job well, you will move from Junior to Senior to Lead. But as AI automates the middle layers of many professions, that ladder is being broken. Progress, by contrast, is active. It involves treating yourself as a product and your career as a series of strategic moves designed to maximise impact and personal satisfaction. This requires a repeatable process for identifying your next step, even when the destination is unclear. You must move beyond 'LinkedIn hacks' and focus on understanding the specific problems, people, and environments that bring out your best work.

The durable career advantage will come from knowing what kinds of problems, people, and environments bring out your best work.

A successful transition requires answering six fundamental questions. You must determine what kind of progress you actually want, what 'quest' your career is currently on, and which specific tasks bring out your best. Most importantly, you must be willing to evaluate tradeoffs. A move that offers more money but less autonomy might be a regression in progress, even if it looks like progression on a CV. The goal is to build a system that allows you to find your bearings every time the labour market throws you a curveball.

The Six Questions of Career Navigation

What kind of career progress do I want?
Which career quest am I on?
Which work brings out my best?
What possible moves should I explore?
Which tradeoffs am I willing to make?
Can I get what I need without leaving?

Consider the professional who, facing a dying startup, assumed his only option was another venture-backed role. By applying a structured evaluation, he discovered that his true desire was for high-impact leadership. This led him to a university entrepreneurship centre—a role his peers would have dismissed as 'non-tech,' but one where he found immediate and profound success. The lesson is that the next right move is rarely the most obvious one; it is the one that matches your internal requirements for meaningful work.

Key Takeaway

Stop chasing the ladder; start designing a system of intentional progress.

05 — Stratechery

The Hardware Pressure Cooker

How Apple and OpenAI are squeezing the Nvidia monopoly

By Stratechery · 5 min read

Editor's note: A brief look at the shifting hardware dynamics in the AI era.

The dominance of Nvidia is being challenged not by a single competitor, but by the strategic shifts of the world's largest platform players. Apple and OpenAI are moving in different directions, but both are exerting pressure on the existing semiconductor hierarchy. Apple's updates to its Mini and Studio lines are increasingly focused on the hardware requirements of local AI execution. Meanwhile, OpenAI's moves suggest a growing desire to control the vertical stack of intelligence, from the model to the specific compute environments it inhabits.

Vertical Integration as Defense

For years, the AI boom has been a windfall for general-purpose GPU manufacturers. However, as the requirements for running large models become more specific, the incentive for companies to build custom silicon grows. Apple has always been a master of this, using its proprietary chips to create a closed ecosystem of performance. As AI becomes a core feature of consumer hardware, Apple's ability to optimize the silicon specifically for their neural engines becomes a massive competitive advantage. They aren't just selling computers; they are selling optimized AI experiences.

Apple and OpenAI represent two different types of pressure on the current compute status quo.

OpenAI, on the other hand, represents the pressure from the top down. As a primary consumer of compute, their massive scale gives them the leverage to demand specialized hardware or even to participate in the design of the next generation of AI-native chips. This creates a pincer movement: Apple optimizes for the edge (the consumer), while OpenAI optimizes for the cloud (the model). Both trajectories point toward a future where the 'general purpose' dominance of current chipmakers is eroded by specialized, vertically integrated players.

Shifting Hardware Dynamics

Apple's focus on local AI execution via optimized silicon
OpenAI's leverage as a massive consumer of cloud compute
The move from general-purpose GPUs to specialized AI accelerators
The increasing importance of vertical integration in the AI stack

The consequence of this shift is a more fragmented and competitive hardware market. While Nvidia currently holds the crown, the strategic necessity of AI is forcing the most powerful companies in the world to become chip designers. The era of relying on a single provider for the building blocks of intelligence is coming to an end.

Key Takeaway

The AI era is driving a shift from general-purpose hardware to specialized, vertically integrated silicon.

06 — The Marginalian

The Suffering of the Storyteller

Why art requires the capacity for loss and the weight of reality

By Maria Popova · 6 min read

Editor's note: A reflection on the essential human conditions that make great art possible.

There is a fundamental difference between a poem written by a machine and one written by a human. An AI can mimic the structure, the meter, and the vocabulary of a master poet, but it cannot capture the essence of the work. This is because a great poem is not merely an arrangement of words; it is a record of experience. Specifically, it is a record of the collision between human desire and the unyielding reality of the world. This collision is what we call suffering, and it is the indispensable ingredient of true art.

The Price of Consciousness

To create something that touches another person, the creator must have lived a life that includes the capacity for loss, for fear, and for the devastation of hope. The philosopher-poet Rainer Maria Rilke argued that to compose something of substance, one must have seen many cities, known the animals, and felt the specific ways in which the world moves. One must have sat beside the dying and experienced the complexities of childhood and adulthood. This is not a romanticised view of pain, but a recognition that consciousness is a full-body phenomenon. Art is the process of transmuting these heavy, often painful experiences into something meaningful.

A poem is made not of language but of experience.

An algorithm is fundamentally incapable of this process because it is spared from failure. An AI follows a programmed plan; if it fails to achieve a goal, it is simply a result of the parameters. It does not feel the sting of a failed aim or the grief of a lost love. Because it cannot suffer, it cannot find the language for the bewilderment of being alive. It can only simulate the appearance of meaning without ever having felt the weight of the reality that makes that meaning necessary.

Requirements for Substantive Creation

The capacity for emotional suffering and loss
Direct experience with the physical and social world
The ability to transmute experience into meaning
A confrontation with the limits of human agency

This realization offers a strange comfort in the age of AI. While machines may eventually master the mechanics of storytelling, they will always be outsiders to the human condition. The very things that make life difficult—the uncertainty, the pain, the unexpected partings—are the very things that make our creative output unique. Art is our way of proving we were here, and that we felt the world in all its complexity.

Key Takeaway

Art is the transmutation of human suffering and experience into meaning, something an algorithm cannot do.